世界銀行-災害對非洲農(nóng)業(yè)的影響：來自微觀數(shù)據(jù)的新證據(jù)（英）-2024.1

Policy

Research

Working

Paper10660e

Imp

acts

of

Disast

ers

on

A

f

rican

A

gricu

lt

u

reNew

Ev

id

encef

rom

Micro-DataPhilip

WollburgYannick

MarkhofomasBentzeGiuliaPonziniDevelop

ment

E

conom

icsDevelop

ment

Data

Grou

pJanu

ary

2024Policy

Research

Working

Paper

10660AbstractDisasters

a?ect

m

illions

of

p

eop

le

each

year

and

causeeconomic

losses

worth

many

b

illions

of

d

ollars

glob

ally.Rep

orting

on

disaster

imp

acts

in

research,

p

olicy,

andnews

p

rimarily

reli

es

on

m

acro

statistics

based

on

disasterinventories.

e

m

acro

statistics

suggest

that

a

relat

iv

elysm

all

share

of

disaster

damages

accrues

in

A

frica.

ispaper,

instead,

uses

detailed

su

rvey

micro-data

f

rom

sixA

f

rican

cou

nt

ries

to

quantify

disaster

damages

in

one

keysector:

crop

agricu

lt

u

re.

e

micro-data

rev

eals

muchhigher

damages

and

m

ore

p

eop

le

a?ected

than

the

m

acrostatistics

would

indicate.

On

average,

36

p

ercent

of

theagricu

lt

u

ral

p

lots

in

the

samp

le

su?er

crop

losses

due

toad

verse

climatic

events.

In

the

cou

nt

ries

and

timep

eriodanalyzed,

these

losses

red

u

ced

total

crop

p

roduction

byanaverage

of

29

p

ercent.Imp

ortantly,

many

of

these

losses

areunderrep

orted

orundetected

in

key

disaster

inventories

andtherefore

elude

macro

statistics.

In

the

case

of

droughts

and?oods,

the

economic

losses

record

ed

in

the

micro-data

are$5.1

b

illion

higher

than

in

the

m

acro

statistics,

a?ecting145

million

to

170

million

p

eop

le,

more

than

fourtimes

asmany

as

the

m

acro

statistics

suggest.

e

d

i?

erence

stemsmostly

f

rom

sm

aller

and

less

sev

ere

but

frequent

ad

verseevents

that

are

not

record

ed

in

disasterinventories.is

paper

is

ap

roduct

of

the

Develop

ment

Data

Group

,

Develop

ment

Economics.

It

is

p

art

of

alarger

e?

ort

by

theWorld

Bank

to

provide

op

en

accessto

its

research

and

make

a

contribution

to

develop

mentp

olicy

discussions

arou

nd

thew

orld

.

Poli

cy

Research

Working

Pap

ersa

re

also

p

osted

on

the

Web

at

http

:///p

rwp

.

e

au

t

hors

maybe

contacted

at

p

wollburg@.e

Policy

Research

Working

Paper

Series

disseminates

the

?ndings

of

work

in

progress

to

encourage

the

exchange

of

ideas

about

developmentissues.

An

objective

of

the

series

is

to

get

the

?ndings

out

quickly,

even

if

the

presentations

are

less

than

fully

polished.

e

papers

carry

thenames

of

the

authors

and

should

be

cited

accordingly.

e

?ndings,

interpretations,

and

conclusions

expressed

in

this

paper

are

entirely

thoseoftheauthors.eydonotnecessarilyrepresenttheviewsoftheInternationalBankforReconstructionandDevelopment/WorldBankanditsa?liatedorganizations,orthoseoftheExecutiveDirectorsofthe

World

Bankorthegovernmentstheyrepresent.ProducedbytheResearchSupport

TeamTheImpactsofDisasterson

AfricanAgriculture:New

EvidencefromMicro-DataPhilip

Wollburg1,2,YannickMarkhof1,3,ThomasBentze1,GiuliaPonzini1JELcodes:Q54,N57,

O13,Q15,C81,

C83Keywords:agriculture,climatechange,disasterrisk,survey

data,Africa,loss

anddamage1

Development

Economics

Data

Group,

World

Bank;

2

Wageningen

University

&

Research;

3

United

NationsUniversity,UNU-MERIT.This

paper

received

funding

support

from

the

50x2030

Initiative

to

Close

the

Agricultural

Data

Gap

and

the

WorldBank

Research

Support

Budget

grant

“On

the

Measurement

of

Agricultural

Productivity

Trends

in

Africa”.

Theauthors

are

grateful

to

Douglas

Gollin,

Erwin

Bulte,

Gero

Carletto,

Ruth

Hill,

Stephane

Hallegatte,

Talip

Kilic,

TravisLybbertandparticipantsoftheCSAE2023Conference,theICASIX

Conference,

theEAAE

2023Congress,andattheWorld

Bank

for

their

comments

and

feedback.

The

findings,

interpretations,

and

conclusions

expressed

in

this

paperare

entirely

those

of

the

authors.

They

do

not

necessarily

represent

the

views

of

the

World

Bank

and

its

affiliatedorganizations,orthoseoftheExecutiveDirectorsoftheWorldBankorthegovernmentstheyrepresent.IntroductionIn

2022,

natural

disasters

led

to

over

$220

billion

in

economic

losses,

affecting

185

million

people.1

Lossesin

2023

are

on

track

to

exceed

the

previous

year’s2

and

large-scale

disasters,

such

as

record

extremeheatwaves,

the

violent

monsoon

in

India,

and

a

prolonged

severe

drought

in

the

Horn

of

Africa

havereceived

widespread

media

and

public

attention.3–6

The

frequency

and

intensity

of

disasters

and

theirimpacts

has

increased

over

the

last

decades,

a

trendthat

is

set

to

continue,

and

likely

accelerate,

due

toclimatechangeandglobal

warming.7–12Reporting

on

disaster

impacts

relies

predominantly

on

macro

statistics.

A

key

data

source

is

theEmergency

Events

Database

(EM-DAT),

which

is

a

publicly

available

global

inventory

of

disaster

impactsthat

is

widely

used

in

media,13

research,14

and

policy

reports,

including

recently

the

World

Bank’s

2023‘Atlas

of

the

Sustainable

Development

Goals’

and

the

Food

and

Agriculture

Organization’s

(FAO)

2021reporton‘Theimpactof

disastersandcriseson

agricultureandfoodsecurity.’15,16Here,

we

offer

a

different

approach

to

studying

disaster

impacts,

based

on

survey

micro-data.

We

quantifythevalueof

crop

productionlossesdueto

adverseclimaticevents

on

morethan

120,000

fieldsacrosssixAfrican

countriesand

study

theimpactsof

these

eventson

African

agriculture,ruralpopulations,and

thenational

economies.

Agriculture

is

a

key

sector,

on

which

many

households

in

the

region

depend

for

theirlivelihoods,

especiallythepoor

and

rural

households.17

Agriculturedependenthouseholdsare

thought

tobe

particularly

at

risk

ofsuffering

theimpactsofclimate

change

andadverseshocks.

Climatechange

andnatural

disasters

are

expected

to

be

especially

severe

in

rural

areas

in

this

region,18,19

while

smallholderagricultural

production

remains

predominantly

rainfed

and

the

adoption

of

drought

or

heat

resistantseeds

or

other

suchclimate-smarttechnologiesis

limited.20We

document

that

crop

losses

due

to

adverse

shocks

are

common

and

costly

both

to

individual

farmhouseholds

and

to

the

economy

at

large,

and

that

farmers

often

suffer

multiple

shocks

in

the

same

season.Taken

together,

production

losses

have

a

substantial

aggregate

impact.

Importantly,

these

events

andtheir

impacts

are

underreported

or

undetected

in

common

macro

data

sources

for

disaster

reporting,suchasEM-DAT.Our

analysis

of

micro-data

offers

an

important

complementary

perspective

to

analyses

based

onaggregate

statistics

derived

from

disaster

inventories.

Aggregate

statistics

are

critical

to

the

study

ofdisaster

impacts,

providing

annual

data

at

a

global

scale.

They

are

less

well-suited

to

capture

thedifferential

impacts

of

disasters

on

different

population

groups,

especiallypoorandvulnerablepeople.21–23

This

is

because

they

account

primarily

for

damages

to

assets

and

losses

in

agricultural

production

whosevalue

is

greater

and

better

documented

among

richer

households

and

in

richer

countries.

For

instance,according

to

the

most

recent

estimates

of

EM-DAT,

about

70%

of

economic

losses

due

to

disastersoccurred

in

the

Americas,

compared

to

just

under

4%

in

Africa.24

A

recent

study

using

the

same

datasourceconcluded

thatdisasterimpactsdonotaffect

poor

peopleasmuchasthe

generalpopulation.25

Incontrast,

evidence

from

survey

micro-data

suggests

that

poorer

households

and

individuals

are

moreexposed

and

less

resilient

to

adverse

climatic

and

environmental

shocks

and

suffer

disproportionatelygreater

well-being

losses

than

better-off

households.18,22,26

Our

analysis

suggests

that

production

lossesdue

to

adverse

climatic

events

are

meaningful

not

only

for

the

well-being

of

low-income

householdsindividually

but,

because

of

how

many

households

are

affected,

they

are

significant

also

for

the

wholeeconomiesof

ourstudy

countriesandon

aglobal

scale.2ResultsCroplossesare

widespreadandsignificantThe

data

used

in

this

analysis

is

from

the

Living

Standards

Measurement

Study-Integrated

Survey

onAgriculture

(LSMS-ISA)

in

Ethiopia,

Malawi,

Mali,

Niger,

Nigeria,

and

Tanzania.

These

data

wereharmonized

across

countries

and

cover

close

to

120,000

fields

on

around

30,000

farms.

The

data

showthat

crop

losses

due

to

disasters

and

adverse

climatic

events

are

widespread

and

significant

in

Africansmallholder

agriculture.

Farmers

report

crop

losses

on

between

11%

(Nigeria

2018/19)

and

90%

of

plots(Niger

2011),

depending

on

country

andyear

(Figure1,

Panel

A

and

Table1).

Overall,

36%ofplots

in

oursamplereportacrop

loss.Farmersreported

losing,on

average,53%oftheirharvestonplotsaffected

bycrop

shocks

(Panel

B

and

Table

1).

Losses

vary

across

countries

and

years,

ranging

from

48%

of

harvest(Ethiopia2018/19)to

71%ofharvest(Niger2011).Disasterlosseshave

also

becomemorecommonovertime

(Table

3).

In

the

11

years

from

2008

to

2019

that

our

dataset

spans,

the

estimated

likelihood

of

aplot

incurring

a

disaster

loss

increased

by

close

to

10

percentage

points.

This

is

seemingly

driven

by

ahigherprevalenceofsmallshocksasthe

estimatedshare

of

harvestlostonplotswithanylossdecreasedby

onaverage1.1percentagepointswith

everyyearstudied.In

aggregate,

crop

losses

due

to

adverse

climatic

events

reduce

the

total

national

crop

production

bybetween3%inNigeria2018-19

and81%inNigerin2011.Atotal

of29%ofpotentialharvestvalue

islostacrossthecountriesandagriculturalseasonsobserved

inourdataset(Figure1,

PanelCandTable4]).3Figure1.PanelAdisplaysthe

prevalenceofcropshockson

plotsacrosscountry-waves.PanelBdisplaysthemeanpercent

ofpotentialharvestvalueloston

plot,by

country-wave,aswellasthefractionof

aggregatepotential

harvestlost(valuedwithcurrent

prices),percountry-waveABCropproductionisimpactedby

multiple

shocksFarmers

face

a

diversity

of

adverse

climatic

shocks.

Multiple

shocks

are

recorded

to

affect

agriculturalproduction

in

each

year

and

across

all

countries

(Table

1).

There

are

also

some

instances

of

multiple

shocks4affecting

thesamefarmin

agiven

agricultural

season

(Table5).

Thisrangesfrom

1.5%

of

farms(Tanzania2014)to

21%of

farms(Ethiopia2018-2019).Overall,droughtisthe

mostcommon

shock,with

22%ofplotsinoursamplerecordingacroplossduetodrought

(Table

1).

One

in

ten

plots

records

losses

due

to

irregular

rains,

meaning

erratic

rainfall

at

unusualtimes

in

the

agricultural

season.

Pests

are

also

widespread

across

our

sample,

affecting

6.3%

ofall

plots.Still,

there

is

substantial

variation

across

countries

and

years.

The

severity

of

the

damages

caused

variesbetween

different

events

(Table

2).

Floods

in

particular

causemore

damage

than

other

shocks,

reducingcrop

production

per

plot

on

average

by

62%.

Losses

from

pests

and

irregular

rains

tend

to

be

smaller.However,thereisagainsomevariation

between

differentcountriesandfarmingenvironments(Table6).Which

shocks

are

the

most

prevalent

varies

also

within

countries.

Figure

2

illustrates

this

for

selectedcountries

and

years,

showing

the

most

reported

events

by

subnational

administrative

divisions.

There

issome

geographical

clustering,

but

we

commonly

see

different

events

accounting

for

most

of

the

impactedplots

in

differentareasof

thesame

countryin

thesame

year.

Thisistrueeven

in

years

with

exceptionallysevere

disasters

such

as

the

droughts

in

Niger

in

2011

and

Ethiopia

in

2015-16

where

the

vast

majoritybutnotallareasof

thecountryrecordeddroughtasthe

primarylossreason.Figure2.Mostcommondisastereventsbyadministrativeunit,selectedcountriesandyears5CroplossesdifferlocallyandbetweenfarmersNotall

farmersand

plotsareequallyaffected.Some

arelesslikelytoexperiencea

loss

even

in

theface

ofan

adverse

climatic

event.

Here

we

show

that

shock

exposure

and

impacts

can

differ

even

betweenneighboring

plots

in

the

same

area.

We

limit

this

analysis

to

droughts.

Given

the

nature

of

droughts,

allplots

in

the

same

small

geographic

cluster

should

be

faced

with

the

same

drought

shock

–but

the

impactsof

that

drought

can

differ.

Indeed,

in

41%

of

the

geographical

clusters

in

our

sample,

some

but

not

all

plotsreportbeingaffectedbyadrought(Table12).

Thisfindingholdsalso

forplotsgrowingthesame

crops.In31%

of

clusters,

some

but

not

all

maize

plots

suffer

drought

losses

(32%

for

sorghum

plots

and

milletplots).

Theresultextendsto

plotswith

thesamecropscultivated

bythesame

households(Table13):

forfarms

thatrecord

adroughtshock

on

one

of

their

maizeplots,

close

to

two-thirdsof

other

maizeplots

onthesamefarmalsorecord

drought-related

crop

losses.These

findings

suggest

that

disaster

impacts

are

highly

localized,

consistent

with

the

high

spatialconcentration

that

meteorological

events

can

have.28

Further,

idiosyncratic

factors,

such

as

landcharacteristics

and

management

practices,

and

happenstance

play

a

role

in

determining

whether

and

howmuch

production

is

affected.

We

find

that

plot

elevation

is

negatively

associated

with

the

likelihood

ofexperiencing

losses

and

the

size

of

the

losses

incurred

(an

effect

almost

twice

as

strong

for

floodscompared

to

other

disasters),

while

smaller

plots

are

less

likely

to

suffer

losses

but

record

higher

losseswhen

they

are

affected

(Tables

7

and

8).

Losses

on

intercropped

plots

are

7.5

percentage

points

lowerthan

on

mono-cropped

plots,

though

intercropped

plots

are

more

likely

to

experience

a

loss

in

the

firstplace

(+3.6

percentage

points).

Plots

farmed

in

more

input

and

technology

intensive

ways

appear

moreresilienttocrop

lossesdue

to

adverseevents.Disaster

exposure

and

impact

also

vary

according

to

who

manages

the

plot.

Plots

managed

by

women

aremore

often

affected

by

disaster

losses

(+2.2

percentage

points)

than

plots

managed

by

men

and

theselossesarealso

largeronaverage

(+4.4

percentage

points;

Table

10).Thesedifferences

are

likely

becauseplots

managed

by

women

are

endowed

and

farmed

differently

than

plots

managed

by

men,

which

in

turnmayfollowfrom

differentialaccessto

inputsandland

between

women

andmen.27Aggregatedatasourcesunderestimate

impactsofextremeevents

oncropproductionHow

dodisaster

impacts

as

captured

in

the

survey

data

compare

toestimates

from

other

commonly

useddata

sources?

Here,

we

contrast

the

results

from

the

survey

microdata

with

publicly

available

estimatesof

disasterimpacts

from

theEmergencyEventsDatabase(EM-DAT).

EM-DATaggregatesreportsfromUNagencies,

governments,

insurance

companies,

research

institutes

and

the

media

into

a

global

inventoryofdisasterimpacts.29

EM-DATisthepreeminentand

only

publiclyavailabledatasourceof

thiskind,usedwidely

in

disaster

reporting

and

research.30

We

focus

on

two

disaster

types,

droughts

and

floods,

andcompare

two

estimates:

the

number

of

people

affected

and

the

total

economic

damages

caused

in

theyears

which

the

survey

micro-data

covers.

We

create

aggregate

figures

from

the

micro-data

usingpopulationsamplingweights.On

both

metrics,

and

for

both

drought

and

flood

impacts,

the

micro-data

estimates

on

average

exceedthe

EM-DAT

estimates,

that

is,

for

years

in

which

there

is

information

from

both

sources,

the

survey

micro-data

find

more

people

affected

and

higher

damages

from

droughts

and

floods

(Figure

3

and

Tables

14

and616).

Moreover,

there

are

many

instances

in

which

the

EM-DAT

records

no

disaster

impacts

at

all.

This

istrue

especially

for

droughts,

where

the

microdata

suggests

that

droughts

are

prevalent

to

some

degreeacross

every

country-year

combination

covered,

while

EM-DAT

records

droughts

affecting

the

populationin

only

a

third

of

cases.

Estimates

of

the

economic

value

of

disaster

impacts

are

mostly

missing

in

the

EM-DAT

data

for

the

study

countries,

even

in

years

when

drought

and

flood

events

were

recorded

to

affectthepopulationinthe

studycountries(Tables15and17).Large,

salient

drought

and

flood

episodes

have

better

coverage

in

the

EM-DAT,

such

as

the

severedroughts

in

Niger

in

201131,32

and

Ethiopia

in

2015-16,33,34

or

the

droughts

and

floods

Malawi

in

2015-16,35,36

which

were

widely

covered

in

international

media

at

the

time.

The

events

that

go

unreported

inEM-DAT

are

smaller,onaverage,in

termsofthepopulationaffected

andthedamagescaused.

However,we

show

that

such

smaller,

under-covered

events

have

substantial

overall

impacts.

More

than

a

fifth

ofthe

population

suffered

production

and

income

losses

in

the

droughts

in

Malawi

in

2009-2010

and

in

Maliin

2074,

according

toour

micro-data

estimates,

while

thereisno

coverage

of

theseevents

in

the

EM-DATfor

thesame

years.

Overall,

weestimatethe

total

numberofpeople

affected

by

droughts

orfloods

in

allinstances

covered

by

the

microdata

is

between

145

million

and

170

million,

more

than

4

times

higher

thanwhatisreportedinthe

EM-DATforthe

sameperiods

andthesameshocks.The

micro-data

analysis

suggests

that

the

aggregate

value

of

the

disaster

impacts

on

crop

production

issubstantial.Forthe

droughtin

Ethiopiain

2015-2016,

themicro-datacroplossestimatesaremuch

largerthan

the

total

economic

damage

reported

in

the

EM-DAT.

For

the

2014

floods

in

Niger,

the

estimatedvalue

ofcrop

losses

exceeds

the

total

damage

reported

in

the

EM-DAT

data

by

almost

USD

78million

(in2022

USD

values).

In

the

other

years

there

is

no

damage

estimate

in

EM-DAT,

but

our

survey

micro-datadocuments

even

some

large

disaster

impacts,

such

as

in

Niger

in

2011

and

Ethiopia

in

2018-2019

withestimated

losses

of

USD

1.6

billion

and

USD

1.4

billion,

respectively.

Taken

together,

we

estimate

thatacross

the

countries

and

years

captured

in

the

microdata,

there

were

USD5.1

billion

in

drought

and

flooddamagesunaccountedforin

theEM-DATdata

(Table

18).Whatexplainsthesediscrepancies?

Disaster

inventoriessuch

asEM-DATandsurvey

microdatadifferinanumber

of

meaningful

ways.

Most

importantly,

disaster

inventories

do

not

measure

shock

impactsthemselves

but

instead

aggregate

data

from

government

sources,

humanitarian

organizations,

the

media,and

others.

They

therefore

rely

on

the

comprehensiveness

and

accuracy

with

which

shocks

due

to

naturalhazards

are

covered

by

one

or

more

of

these

sources.30,37

Less

salient

events

as

well

as

those

affectingmarginalized

population

groups

are

less

likely

to

be

reported

on

and

less

likely

to

have

detailedinformation

onthe

affectedpopulation

oreconomicandwelfareimpacts.37–40

ThisisparticularlyacuteinLMICs

where

the

density

of

information

for

disaster

repositories

to

draw

on

is

much

lower

and

a

largeshareof

damagesis

uninsured.30,37ShocksinLMICsin

generaland

smallerevents

(interms

ofintensityorthe

population

affected)

in

particular

are

more

likely

to

have

incomplete

or

inaccurate

information

indisaster

repositories

or

are

not

covered

at

all.30,37,41,42

Microdata

such

as

the

LSMS-ISA

measure

shockimpacts

on

smallholder

farmers

where

they

occur

byasking

farmers

directly.

They

therefore

do

not

sufferfrom

the

same

limitations

regarding

the

recording

of

smaller,

less

salient,

or

more

localized

shocks

andtheirimpactsas

disasterrepositories.Smaller

shocks

or

adverse

climatic

events

may

not

be

considered

disasters

as

disasters

suggest

a

minimumlevel

of

severity.

For

an

event

to

be

recorded

in

the

inventory,

the

EM-DAT

requires

a

minimum

of

100people

to

be

affected

(injured,

homeless,

in

need

of

immediate

assistance)

or

an

official

declaration

of7emergency

or

appeal

for

international

assistance

–

arguably

a

sensible

set

of

criteria

for

a

disasterinventory.

Not

all

events

recorded

in

the

micro-data

meet

these

requirements.

Importantly,

the

eventsrecorded

in

the

micro-data

have

substantial

impacts

on

the

livelihoods

of

farmers

and

the

economies

ofthestudycountries.At

the

same

time,

micro-data

has

drawbacks

and

limitations.

First,

it

is

rare

that

microdata

in

low-

andmiddle-income

countries

are

available

annually,

with

surveys

typically

implemented

every

few

years.Shock

coverage

and

detail

depend

on

the

survey

design,

which

typically

differs

from

country

to

country.Finally,microdatadoesnotprovidethesamecross-countrycoverageasdisasterrepositories.Withtheselimitations,themicrodatanaturallyalso

providesanincompletepicture(seediscussion

inAppendixA).8Figure3.ComparisonofshockprevalenceandimpactbetweenEM-DATandLSMS-ISAdata.ABPercentaffectedinLSMS(with95%CI)TotaldamagesinLSMS(with95%CI)PercentaffectedinEM-DATTotaldamagesinEM-DATCDPercentaffectedinLSMS(with95%CI)TotaldamagesinLSMS(with95%CI)PercentaffectedinEM-DATTotaldamagesinEM-DATNote:PanelAdisplaysacomparisonofthetotalestimatedindividualsaffectedbydroughts

betweenEMDAT(in

blue)andLSMS-ISAdata

(inorange),whilepanelBshowsacomparisonof

theestimateddamages(inmillions

of2022dollars),inyearswheredamagescouldbeestimatedintheLSMS-ISAsurveys.

PanelCdisplaysasimilarcomparisonforfloods,in

yearswherefloodsarelistedasapotentialshockin

theLSMS-ISAdata,while

panelDshowsacomparisonofestimateddamagesfromfloods.Confidenceintervals

for

panels

BandDwerecalculatedbeforelog-transformation,andarehenceasymmetricallysituatedaroundlog-scaledpointestimates.DiscussionWe

explore

the

crop

production

impacts

of

adverse

climatic

events

on

120,000

fields

on

30,000smallholder

farms

in

Sub-Saharan

Africa.

Smallholderagriculture

is

ofspecial

interest

for

achieving

SDGs1and2asitremainsthe

primarymeansoflivelihoodformanyof

theworld’spoor.43Our

findings

generate

new

insights

and

advance

our

understanding

of

the

disaster

risks

and

losses

thatsmallholder

farmers

face.

Other

studies

have

investigated

the

vulnerability

of

smallholder

farmers

to9disasters

and

environmental

shocks.44–47

These

stud